dart2js TypeImpl overrides operator ==

[floitschG]

The equality operator is overridden for dart2js' Type implementation TypeIpml.

This means that it cannot be used in switch statements or as keys in constant maps.

See issue #15295 and issue #17123.

[floitschG]

It's not clear if this is a bug or not. The VM does not override it, and this leads to incompatibility between VM and dart2js.

[rakudrama]

Marked this as being blocked by #21553.

[floitschG]

Issue #23009 has been merged into this issue.

---

cc @karlklose.
cc @johnniwinther.

[jakemac53]

Just ran into this as well. It seems pretty clear to me that this is a bug because it causes programs which are completely valid in the vm to fail in dart2js.

[lexaknyazev]

Apparently, this is somehow fixed with CFE.

[sigmundch]

Note - the CFE is missing some compile-time errors. I believe this is not actually fixed, since we still override == in our implementation of Type, so I expect this error should be visible again in the near future.

[sigmundch]

I just filed #32557 reporting why the CFE is not handling this error at this time. The underlying issue in dart2js remains, so I recommend not using Types as const keys or values on switch cases at this time, since it will be an error again soon.

[rakudrama]

The language specification currently treats type literals specially:

"Every instance of type Type which was originally obtained by evaluating a constant type literal
has a primitive operator ==."

I read this as permitting TypeImpl to override ==, but the implementation is known to the compiler and runtime, just like the implementation of Object.== is known to the compiler and runtime, so constant type literals are permitted as case expressions and constant map keys.

[sigmundch]

We are very close to be able to mark this issue as fixed.

The new RTI is now enabled by default since 2.8.0-dev.1.0. This already makes switch cases work properly because our types are properly canonicalized. One issue remains is that we still override operator== in _Type (the equivalent of TypeImpl in the new RTI), but given our canonicalization rules we should be able to remove that.

@rakudrama - @fishythefish and I assume that's safe to remove at this point, correct?

[sigmundch]

== was removed in 3c2a6c2